[PATCH net-next] net: add the ifindex for trace_net_dev_xmit_timeout

[PATCH net-next] net: add the ifindex for trace_net_dev_xmit_timeout

[Tonghao Zhang]

In a multi-network card or container environment, provide more accurate information.

268.148239: net_dev_xmit_timeout: dev=eth0 ifindex=6 driver=virtio_net queue=10
274.292271: net_dev_xmit_timeout: dev=eth0 ifindex=6 driver=virtio_net queue=10
277.876231: net_dev_xmit_timeout: dev=eth1 ifindex=7 driver=virtio_net queue=10
279.924230: net_dev_xmit_timeout: dev=eth2 ifindex=8 driver=virtio_net queue=10
279.924233: net_dev_xmit_timeout: dev=eth0 ifindex=6 driver=virtio_net queue=10
282.996224: net_dev_xmit_timeout: dev=eth1 ifindex=7 driver=virtio_net queue=10

Cc: Eran Ben Elisha <eranbe@mellanox.com>
Cc: Jiri Pirko <jiri@mellanox.com>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Tonghao Zhang <tonghao@bamaicloud.com>
---
 include/trace/events/net.h | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/include/trace/events/net.h b/include/trace/events/net.h
index d55162c12f90..49f9fce69dd7 100644
--- a/include/trace/events/net.h
+++ b/include/trace/events/net.h
@@ -107,16 +107,19 @@ TRACE_EVENT(net_dev_xmit_timeout,
 		__string(	name,		dev->name	)
 		__string(	driver,		netdev_drivername(dev))
 		__field(	int,		queue_index	)
+		__field(	int,		ifindex		)
 	),
 
 	TP_fast_assign(
 		__assign_str(name);
 		__assign_str(driver);
 		__entry->queue_index = queue_index;
+		__entry->ifindex = dev->ifindex;
 	),
 
-	TP_printk("dev=%s driver=%s queue=%d",
-		__get_str(name), __get_str(driver), __entry->queue_index)
+	TP_printk("dev=%s ifindex=%d driver=%s queue=%d",
+		__get_str(name), __entry->ifindex,
+		__get_str(driver), __entry->queue_index)
 );
 
 DECLARE_EVENT_CLASS(net_dev_template,
-- 
2.34.1

Re: [PATCH net-next] net: add the ifindex for trace_net_dev_xmit_timeout

[Jakub Kicinski]

On Tue, 21 Oct 2025 17:19:00 +0800 Tonghao Zhang wrote:
> In a multi-network card or container environment, provide more accurate information.

Why do you think that ifindex is more accurate than the name?
Neither is unique with netns..
-- 
pw-bot: cr

Re: [PATCH net-next] net: add the ifindex for trace_net_dev_xmit_timeout

[Tonghao Zhang]

> On Oct 22, 2025, at 08:10, Jakub Kicinski <kuba@kernel.org> wrote:
> 
> On Tue, 21 Oct 2025 17:19:00 +0800 Tonghao Zhang wrote:
>> In a multi-network card or container environment, provide more accurate information.
> 
> Why do you think that ifindex is more accurate than the name?
> Neither is unique with netns..
I thought ifindex was globally unique, but in fact, different namespaces may have the same ifindex value. What about adding the ns.inum inode ?

> -- 
> pw-bot: cr
> 
> 

Re: [PATCH net-next] net: add the ifindex for trace_net_dev_xmit_timeout

[Ido Schimmel]

On Wed, Oct 22, 2025 at 04:31:34PM +0800, Tonghao Zhang wrote:
> > On Oct 22, 2025, at 08:10, Jakub Kicinski <kuba@kernel.org> wrote:
> > 
> > On Tue, 21 Oct 2025 17:19:00 +0800 Tonghao Zhang wrote:
> >> In a multi-network card or container environment, provide more accurate information.
> > 
> > Why do you think that ifindex is more accurate than the name?
> > Neither is unique with netns..
> I thought ifindex was globally unique, but in fact, different namespaces may have the same ifindex value. What about adding the ns.inum inode ?

The netns cookie is probably better. See this thread:

https://lore.kernel.org/netdev/c28ded3224734ca62187ed9a41f7ab39ceecb610.camel@fejes.dev/

You can also retrieve it by attaching a BPF program to the tracepoint.

Re: [PATCH net-next] net: add the ifindex for trace_net_dev_xmit_timeout

[Tonghao Zhang]

> On Oct 22, 2025, at 18:45, Ido Schimmel <idosch@idosch.org> wrote:
> 
> On Wed, Oct 22, 2025 at 04:31:34PM +0800, Tonghao Zhang wrote:
>>> On Oct 22, 2025, at 08:10, Jakub Kicinski <kuba@kernel.org> wrote:
>>> 
>>> On Tue, 21 Oct 2025 17:19:00 +0800 Tonghao Zhang wrote:
>>>> In a multi-network card or container environment, provide more accurate information.
>>> 
>>> Why do you think that ifindex is more accurate than the name?
>>> Neither is unique with netns..
>> I thought ifindex was globally unique, but in fact, different namespaces may have the same ifindex value. What about adding the ns.inum inode ?
> 
> The netns cookie is probably better. See this thread:
> 
> https://lore.kernel.org/netdev/c28ded3224734ca62187ed9a41f7ab39ceecb610.camel@fejes.dev/
That's a good idea, it's helpful to me.
> 
> You can also retrieve it by attaching a BPF program to the tracepoint.
>